Chris-davis-thor/motioncapture

Fight Scene Action Video Dataset (samples)

A VideoFolder-format dataset of fight / martial-arts action clips, provided by Thordata. Each clip is annotated with a temporal action breakdown, per-segment descriptions, an English caption, and ffmpeg-measured technical metadata.

Layout (HuggingFace VideoFolder)

data/
├── metadata.jsonl          # one row per video; "file_name" links to the clip
└── train/
    ├── clip_01.mp4
    └── ... clip_12.mp4

Load

from datasets import load_dataset
ds = load_dataset("videofolder", "<your-username>/fight-scene-action-video")
print(ds["train"][0]["caption"])
print(ds["train"][0]["json"]["action_labels"])

Fields

Field	Type	Description
file_name	string	Relative path to the video file
id	string	Clip identifier
provider	string	Data provider (Thordata)
title	string	Short English title
caption	string	One-sentence English description
style	string	Fight style (traditional kung fu / Wing Chun / street brawl / ground grappling / weapon duel)
setting	string	Scene / environment
fighters	string	Who is fighting
camera	string	Camera / cinematography note
action_labels	list<string>	Coarse action tags for classification
segments	list<dict>	Temporal breakdown: start, end (seconds), action, description
duration_s	float	Duration (seconds)
width, height	int	Resolution
fps	float	Frames per second
num_frames_est	int	Estimated frame count
has_subtitle_or_watermark	bool	Always false for these clips

Coverage (12 clips)

Outdoor changshan kung-fu duel, street mob brawl, superhero-style ground control, night interior hand-to-hand, green-lit interior standoff, woman fighting in a hardware store, indoor bladed-weapon duel, Wing Chun trapping & chain punches, Wing Chun palm blocks / punches / kick, Wing Chun bridge control & throw, one-vs-many hall brawl with dagger, broadsword weapon duel.

Citation

@misc{fight_scene_action_video_samples,
  title    = {Fight Scene Action Video Dataset (samples)},
  author   = {Thordata},
  year     = {2026}
}